Enabling Bit-by-Bit Backscatter Communication in Severe Energy Harvesting Environments

Transcription

1 Enabling Bit-by-Bit Backscatter Communication in Severe Energy Harvesting Environments Pengyu Zhang Deepak Ganesan! University of Massachusetts Amherst

2 Network Stacks on Energy Harvesting Sensors Micro-powered sensors Glucose sensor in bloodstream Vital signs bandaid In-ear molecular sensor Google contact lens Epidermal electronics

3 Network Stacks on Energy Harvesting Sensors Micro-powered sensors Glucose sensor in bloodstream Vital signs bandaid In-ear molecular sensor Google contact lens Epidermal electronics Energy Harvester Micro-solar cells Vibration Thermal gradient Wireless signal Biofuel cells

4 Network Stacks on Energy Harvesting Sensors Network stack PHY Link MAC Transport APP Micro-powered sensors Glucose sensor in bloodstream Vital signs bandaid In-ear molecular sensor Google contact lens Epidermal electronics Energy Harvester Micro-solar cells Vibration Thermal gradient Wireless signal Biofuel cells

5 RF Harvesting and Backscatter Communication Backscatter reader Backscatter communication Sensor RF harvesting Backscatter radios enable RF energy harvesting as well as ultra low power wireless communication.

6 Big Gap between Operational and Comm Range 3.6 feet Dewdrop [NSDI 2011] RF harvesting Wireless communication

7 Big Gap between Operational and Comm Range 3.6 feet Dewdrop [NSDI 2011] Flit [MobiSys 2012] EPC Gen feet RF harvesting Wireless communication BIG gap! Communication Range Why a big gap between the operational range and the comm range?

8 Energy Harvesting Systems Macro energy harvesting Micro energy harvesting Huge energy buffer Long time scale Tiny energy buffer ~ms time scale TinyOS TinyOS TinyOS TinyOS Day Night ms ms Tiny amount of energy accumulated for a single discharge cycle.

9 Why is operational range limited? Dewdrop [1] Mementos [3] EPC Gen 2 Flit [2] Key - task executed needs to complete within a single discharge cycle. Energy [1] Dewdrop. NSDI [2] Flit. MobiSys [3] Mementos. ASPLOS 2011.

10 Why is operational range limited? Dewdrop [1] atomic unit handshake packet Atomic units cannot fit into a single discharge cycle. Insight - drive the atomic unit down to the smallest unit! [1] Dewdrop. NSDI 2011

11 Transmit the Smallest Atomic Unit Packet - atomic unit of a network stack IP Payload Energy 1 bit sleep 1 bit sleep 1 bit sleep 1 bit - atomic unit of a network stack

12 Operational Range Benefit of Bit-by-bit Transmission 3.6 feet Dewdrop [1] BIG increase! 18 feet TX bit-by-bit 18.6 feet Communication Range Transmitting 1 bit each time pushes the operational range to the extreme. [1] Dewdrop. NSDI 2011

13 Throughput suffers for bit-by-bit transfer Transmit 1 bit in each discharge cycle 3.6 feet How 100 bps Throughput suffers at close distance 18 feet 80 bps How to achieve maximum communication throughput while still maintain the capability of operating at the maximum range?

14 Auto tuning µframe controller Goal - maximize communication throughput while maintain the capability to operate at the maximum range. Decision - Decide which parameters to tune for maximizing communication throughput as well as operational range. sleep interval µframe size Tuning - How to tune each parameter for throughput and operational range optimization? Energy Harvesting Rate (uw) Sleep Time (seconds) TinyOS Signal Strength transmitted signal Transmission Time (ms) decoding fails

15 Which parameters to tune for optimal throughput? What parameters to tune cheaply? Voltage Dewdrop [1] Mementos [2] Expensive for micro powered devices! 1 voltage measurement = 27 bits TX [1] Dewdrop. NSDI [2] Mementos. ASPLOS 2011.

16 Which parameters to tune for optimal throughput? What parameters to tune cheaply? Voltage Dewdrop [1] Voltage Mementos [2] Time Expensive for micro powered devices! 1 voltage measurement = 27 bits TX Sleep time TX bits Measuring bits and time is cheaper than measuring voltage [1] Dewdrop. NSDI [2] Mementos. ASPLOS 2011.

17 How to choose the sleep time? Voltage Long sleep time = more energy buffered Sleep time TX bits

18 How to choose the sleep time? igh impact. Figure 3(a) shows the empirharvesting rate as we vary the amount of the node replenishes energy between two The results are counter-intuitive while ct more Voltageenergy to be harvested over time, rate drops to zero for longer sleep duraation can be explained analytically 160 by 140 capacitors Sleep time buffer TX bits energy. The 120 charga capacitor follows its charging equation t s /t), where t s is the sleep time, t is the constant, and V max is the maximum volte capacitor Gradient 20 searchcan be charged under0 the curconditions. Its energy harvesting rate folon: H = C Vmax 2 t 1 (1 e ts/t )e ts/t. esting conditions are constant (i.e. V max Energy Harvesting Rate (uw) 180 ing, resulting In order to at how a back provides powe nication. The be reflected b its own inform toggling the s shown in Figu shared by diff the incoming p Close to zero device while communicatio of the energy circuit, which C when the vo Optimal Sleep Time (seconds)

19 How to choose the number of bits in a µframe? Voltage Deplete charged energy and transmit as much as possible? Sleep time TX bits

20 Voltage How to choose the number of bits in a µframe? Sleep time TX bits Signal Strength Surprising SNR degradation with increasing µframe length Transmission Time (ms) antenna energy buffer Less RF power is reflected when the local energy becomes low. transistor matching circuit C system

21 Tuning µframes for the sensor-to-reader and readerto-sensor links µframe sleep µframe sleep Auto tuning µframe controller Sensor-to-reader comm link

22 Tuning µframes for the sensor-to-reader and readerto-sensor links µframe sleep µframe sleep Auto tuning µframe controller Sensor-to-reader comm link Available energy on the sensor? Reader-to-sensor comm link

23 Remote interrupt - interrupt reader-to-sensor stream Backscatter reader s signal

24 Remote interrupt - interrupt reader-to-sensor stream Backscattered sensor signal

25 In-band Remote interrupt interrupt the stream when sensor s energy is low Remote interruption Independent communication: 1. The backscatter reader turns on and off the carrier wave. 2. The sensor detunes and tunes the carrier wave.

26 How to interleave µframes across sensors? MAC for interleaving µframes Why should we interleave? Channel efficiency Challenges How to interleave? token bucket Auto tuning µframe controller Selecting the sleep time Auto tuning for the reader-to-sensor TinyOS comm link TinyOS Selecting the num of TX bits Implementation - trim the TinyOS overhead of each µframe

27 Interleaving sensors on the packet level Packet from sensor 1 Packet from sensor 2

28 Interleaving sensors on the packet level Packet from sensor 1 Packet from sensor 2 µframe sleep µframe sleep µframe µframe sleep µframe packet from sensor 1 packet from sensor 2

29 Interleaving sensors on the µframe level µframe1 µframe2 µframe1 sleep µframe2 µframe1 sleep µframe2 interleaving at µframes level

30 Token bucket based scheduling algorithm Estimated sleep time and TX bits Token bucket scheduler µframe1 µframe2 µframe1 sleep µframe2 µframe1 sleep µframe2 interleaving at µframes level

31 System Implementation WISP/Moo 10m USRP 1 20m

32 Distance (feet) Benefit on the Operational Range Dewdrop QuarkNet 3.5x [1] Dewdrop. NSDI 2011 RF Power (dbm)

33 Scheme Big Increase on Communication Throughput 20 Throughput (kbps) x 1.2x 1.14x 5.8x 0 [1] Dewdrop. NSDI 2011 Dewdrop +adaptive uframe +adaptive SNR QuarkNet

34 Throughput (kbps) Throughput of Reader-to-sensor Comm Link QuarkNet EPC Gen 2 9x Distance (feet)

35 Conclusion QuarkNet enables a network stack to seamlessly scale down to severe energy harvesting conditions. Key contribution is a novel packet fragmentation abstraction that can scale down to frame sizes as small as a few bits while adapting to harvesting and network conditions. QuarkNet addresses a fundamental limitation of backscatter, which is operating range, and simultaneously improves throughput. - Maximum operational range 21 feet, 3.5x higher than Dewdrop. - Maximum throughput 18 kbps, 5.8x higher than Dewdrop.

36 Backup Slides

37 Results: selecting optimal sleep time 1 Normalized Throughput lux at 1ft 350lux at 6ft 0 150lux at 6ft Sleep Time (log(ms))

38 Throughput of Interleaving Sensors Throughput (kbps) Individual Interleaving 5.4x RF Power (dbm)

39 Throughput (kbps) teristic of parameters ured harvesting ratethe asharvesting we vary thesource, amountsystem of In order a surprisingly high impact. Figure 3(a) shows ich the node replenishes energy between two at howthea em b ns. The ically resultsmeasured are counter-intuitive while harvesting rate as we vary the amou provides po How to choose the sleep time? xpect more to bethe harvested over time, energy timeenergy for which node replenishes between nication. T 25 Voltage Empirical Curve ng rate transmissions. drops to zero for durabe reflected Thelonger resultssleep are counter-intuitive w Theoretical Curve 20 its ownover info one might expect more energy to be harvested Optimal 15 ervationthecan be explained analytically toggling harvesting rate drops to zero by for longer sleepthed 10 Close to zero how capacitors buffer energy. The chargshown in Fi tions. Sleep time TX bits of a capacitor its charging by d This follows observation can 5 beequation explained shared analyticall 0 e ts /tlooking ), whereatts is the capacitors sleep time, t is the the incomin how buffer energy The 18 20c Gradient ime constant, and Vmax the maximum voltsleep Time (seconds) device whi search ing process of ais capacitor follows its charging equ h the capacitor can under the communica V = Vmax (1be charged e ts /t ), where ts iscurthe sleep time, t i ing conditions. Its energy harvesting rate fol-is the of the ener RC circuit time constant, and V maximum max Sleep time should be chosen for 2 1 t /t t /t s s uation: H = to C V t capacitor (1 e optimizing )ebeenergy. harvesting circuit, whi max rate. age which the can charged under the harvesting conditions constant (i.e. Vmax harvesting C when the rent harvestingareconditions. Its energy rate

40 Trim the overhead of each transmission Remove the pilot tone of each frame... TinyOS... pilot tone preamble payload data

41 Trim the overhead of each transmission Fixed Cost (us) QuarkNet Overhead (us) TX to inactive 9.9 interrupt config inactive to TX 47.5 handle interrupt 9.3 RX to TX 4.08 frame adaptation 24.3 sleep to wakeup 9.83 voltage detection us > 47.18us

42 Application Task fragmentation can be used for operating a micro powered image sensor and transmitting the sensor data back via backscatter. 1 pixel sleep 1 pixel 1 pixel sleep 1 pixel 1 pixel control signal sleep amplifier ADC sleep

43 Decoding variable length µframes Amplitude 1st falling edge last falling edge